Ephemeral rivers in dryland regions exhibit a high interannual variability of streamflow regime, mainly dominated by floods. In these environments, floods are a water resource and a potential hazard with important socioeconomic implications. The Fish River (86,600 km²) is the largest ephemeral stream in Namibia and, recently, also the focus of new development plans, including construction of the largest dam in Namibia. The hydrological analysis to support decisions implies large uncertainties owing to spatial-limited and short hydrological records (since 1962). Here we investigate the current and past patterns of extreme floods combining instrumental, historical, and palaeoflood records. Palaeoflood studies were performed at two reaches with preserved sedimentary evidence: at the upper sector (Vogelkranz) upstream of Hardap Dam (~13% catchment area) and in the lower part of the river, in the Fish River Canyon National Park (70% catchment area). In the Hardap reach, the palaeoflood record identified at least eight large floods during the last 350 years, with the largest flood reaching a minimum discharge of 4800 m3 s−1 (150-year return period). In the Fish River Canyon reach, the sedimentary record shows at least 12 large floods over the last 400 years, the largest with an estimated minimum discharge of 8700 m3 s−1. The elevation of alluvial surfaces without flood evidence provided an upper bound for flood stages, associated with discharges of 6400 m3 s−1 and 16,140 m³ s−1 for the upper and mid-lower Fish River respectively. In the upper reach, the flood frequency analysis (FFA) combining systematic and palaeoflood data provided lower discharges (~25%) of the flood quantiles than the ones using only systematic data sets. In the Fish River Canyon reach, incorporation of the palaeoflood data into the FFA results in slightly higher values in the magnitude of the higher flood quantiles (~4–7%). The FFA analysis using an upper limited lognormal distribution function (LN4) fitted with palaeoflood data shows a good performance with a slow behaviour approaching the upper limit. Our flood frequency results suggest that the Hardap Dam should increase the spillway capacity and safety check flood of the original design in order to satisfy the dam safety criteria. However, the projected reevaluation figures calculated from conventional hydrological methods results in an overestimation of the safety floods according to our estimations. Keywords: Palaeoflood, Hydrology, Ephemeral rivers, Flood frequency analysis.